llms.txt and Schema for AI Crawlers: What Actually Moves Citations

Teams keep asking whether adding an llms.txt file or more schema will get them cited by AI engines. The honest answer: these are enablers, not causes. They make your best content readable and extractable for AI crawlers, which matters, but a model still has to find something worth quoting. Here is what each does, what it does not do, and the order to do them in.

What llms.txt actually is

llms.txt is a markdown file you place at the root of your domain (yoursite.com/llms.txt). It is a curated map: a short description of who you are, plus links to the pages you most want AI systems to read and cite. Think of it as a hand-built index for language models, the way robots.txt and sitemap.xml serve traditional crawlers.

It does two useful things. It signals which content is canonical and answer-ready, so a crawler does not have to guess. And it gives smaller or newer sites a clean entry point instead of hoping a model stumbles onto the right page.

What it does not do is force citations. Publishing the file does not put you in answers. It just makes sure that when a model does look, it finds your strongest, clearest pages first. If you want one without hand-writing the markdown, the free llms.txt generator builds a valid file from your site in a couple of minutes.

What schema does for AI extraction

Schema, or structured data, is machine-readable labeling (usually JSON-LD) that tells a crawler what a piece of content means: this is an FAQ, this is the question, this is the answer, this is the author, this is the organization.

For AI engines, schema lowers extraction risk. When a model can see that a block is explicitly marked as an answer to a specific question, it can lift that answer with more confidence than guessing from raw prose. FAQPage and Article schema are the two that pull the most weight for AEO, because they map directly onto how people query AI assistants: in questions.

The table below sorts what each layer is for.

The mistake: treating these as the whole job

The pattern we see most: a team adds llms.txt and schema, waits a month, and sees no movement in AI answers. The fix was never going to come from markup alone. Schema labels content; it does not create a quotable claim. If the underlying page does not open with a tight, standalone answer, there is nothing clean for the model to lift.

That is why the structural work comes first. The page has to answer the real question in the first 40-60 words, then back it with clear H2 sections, a table where it fits, and real entity clarity. We break the full pattern down in what content structure actually gets cited by AI assistants. Schema and llms.txt then make that good content easy to find and safe to extract.

The right order of operations

Run it in this sequence and each step pays off the next.

1. Identify the real questions buyers ask AI engines in your category, and where competitors already get cited. The method is in our guide to getting cited by AI engines.
2. Write answer-first content for those exact questions: capsule, structure, table, FAQ.
3. Add FAQPage and Article schema so the answers are labeled and extractable.
4. Publish an llms.txt that points to those pages.
5. Keep the feed fresh. Models favor sources that publish answer-shaped content consistently, which is the job the AI Feed Engine runs on autopilot.
6. Measure citation share monthly so you know what is working.

You can see how the whole engine fits together on the how OnlyAEO works page, and what it costs to run on the OnlyAEO pricing page. For a worked example of this exact sequence moving citations, the FastTrackr AI case study walks through the before and after.

Frequently asked questions